Retired battery screening method, system and equipment based on capacity increment curve prediction and medium

Abstract

The invention discloses a retired battery screening method, system and equipment based on capacity increment curve prediction and a medium, and relates to the technical field of battery application. The method comprises the following steps: firstly, acquiring local charging data, including voltage and capacity, of a target battery, and performing discretization processing to obtain preprocessed data; then, a voltage-capacity curve prediction model based on a convolutional neural network is utilized to predict a complete charging curve of the preprocessed data, and a capacity increment curve is generated; and then, extracting characteristic data of the capacity increment curve, including a left peak height, a right peak height and corresponding voltage, as clustering indexes to screen out-of-service batteries. And finally, determining a judgment result of the target battery according to the feature data, and realizing accurate screening and evaluation of the battery. According to the method, the clustering feature acquisition time can be greatly shortened, the clustering efficiency is improved, and rapid screening of the retired batteries is realized.

Description

Technical Field

[0001] The present invention relates to the field of battery application technology, and in particular to a retired battery screening method, system, device and medium based on capacity increment curve prediction. Background Art

[0002] Due to the large number of retired batteries and the large differences in operating conditions, there is a high inconsistency between parameters such as capacity, internal resistance and voltage of retired batteries. The consistency problem of retired batteries must be solved before they are reused. Otherwise, not only will the module capacity be restricted by the battery with the lowest capacity, affecting its electrical performance, but it may even cause some batteries in the module to be overcharged or over-discharged, leading to thermal runaway, which poses a serious safety hazard. Therefore, it is necessary to select indicators that can reflect the consistency of batteries for clustering and grouping. The key to clustering retired batteries is the selection of clustering features. There are many aging mechanisms in batteries, namely the loss of lithium ions (LLI), the loss of active materials (LAM) and the thickening of the SEI film, which ultimately manifests as an increase in the internal resistance of the battery and a decrease in capacity.

[0003] As significant aging characteristics, the remaining capacity and internal resistance are the most common clustering indicators in the cascade utilization of retired batteries. However, the capacity and internal resistance of lithium batteries cannot usually be directly obtained online, and the acquisition of capacity requires complete charging and discharging under specified conditions, while the acquisition of internal resistance requires offline AC impedance testing. BMS cannot directly measure the internal resistance online. Therefore, the current efficiency of retired battery feature clustering is not high, which leads to the problem of low battery screening efficiency. Summary of the invention

[0004] The purpose of the present invention is to provide a retired battery screening method, system, device and medium based on capacity increment curve prediction, which can greatly shorten the clustering feature acquisition time, improve clustering efficiency, and realize rapid screening of retired batteries.

[0005] To achieve the above object, the present invention provides the following solutions:

[0006] A retired battery screening method based on capacity increment curve prediction, comprising:

[0007] Acquiring local charging data of a target battery; the local charging data including voltage and capacity;

[0008] Discretizing the charging data to obtain preprocessed data;

[0009] Inputting the preprocessed data into a voltage-capacity curve prediction model to predict a complete charging curve, and obtaining a capacity increment curve based on the predicted complete charging curve; the voltage-capacity curve prediction model is constructed based on a convolutional neural network;

[0010] Extract characteristic data of the capacity increment curve, and use the characteristic data as a clustering index to screen retired batteries to determine the discrimination result of the target battery; the characteristic data includes a left peak height, a right peak height and a corresponding voltage.

[0011] Optionally, the voltage-capacity curve prediction model consists of a ten-layer network structure, wherein the first layer is the input layer, which receives the discretized data; the second layer is the convolution layer, which contains two filters, each filter contains 16 convolution kernels with a structure of 3×3, and the activation function is ReLU; the third layer is the pooling layer, the size of the pooling window is 3×3, and the pooling method is maximum pooling; the fourth, fifth, sixth and seventh layers are also convolution layers and pooling layers, the same as the second and third layers; the eighth layer is a fully connected layer with 140 neuron nodes, and the activation function is ReLU; the ninth layer is the Dropout layer with a dropout rate of 0.2, which is used to randomly select neurons to zero to suppress overfitting; the tenth layer is the output layer, which has a total of 140 neurons, corresponding to 140 sampling points of the complete curve, and is used to output the predicted complete charging curve.

[0012] Optionally, the preprocessed data is input into a voltage-capacity curve prediction model to predict a complete charging curve, and the specific process includes:

[0013] The preprocessed data is used as the input matrix I of CNN, and the input matrix I includes two rows of voltage and capacity, wherein the sequence voltage with equal step size represents the sampling position of the discretized capacity, the sequence voltage of different curves is the same, and the capacity is different between different batteries and different curves due to aging. The input matrix I is obtained by subtracting the capacity corresponding to the moment V(t-1) from the capacity sequence, setting the initial capacity value to zero, and starting to calculate the charging capacity until the moment V(t+m), 1≤m≤N-1, 2≤t≤N+1-m, N represents the total length of the charging curve, and m represents the length of the local charging curve.

[0014] Optionally, the input matrix I is expressed as:

[0015]

[0016] Among them, V represents voltage and Q represents capacity.

[0017] Optionally, extracting characteristic data of the capacity increment curve and using the characteristic data as a clustering index to screen retired batteries to determine a discrimination result of the target battery specifically includes:

[0018] Extracting features from the capacity increment curve to obtain feature data corresponding to the curve;

[0019] Clustering is performed based on the characteristic data, and combined with a t-test, outliers in the clustering results are eliminated to determine the discrimination results of the target battery and complete the retired battery screening.

[0020] The present invention also provides a retired battery screening system based on capacity increment curve prediction, comprising:

[0021] A data acquisition unit, used to obtain local charging data of a target battery; the local charging data includes voltage and capacity;

[0022] A preprocessing unit, used for discretizing the charging data to obtain preprocessed data;

[0023] A model prediction unit, used for inputting the preprocessed data into a voltage-capacity curve prediction model to predict a complete charging curve, and obtaining a capacity increment curve based on the predicted complete charging curve; the voltage-capacity curve prediction model is constructed based on a convolutional neural network;

[0024] The retired battery screening unit is used to extract the characteristic data of the capacity increment curve, and use the characteristic data as a clustering index to screen the retired batteries to determine the discrimination result of the target battery; the characteristic data includes the left peak height, the right peak height and the corresponding voltage.

[0025] The present invention also provides an electronic device, including a memory and a processor, wherein the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute the retired battery screening method based on the capacity increment curve prediction mentioned above.

[0026] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the retired battery screening method based on capacity increment curve prediction as described above.

[0027] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0028] The present invention discloses a method, system, device and medium for screening retired batteries based on capacity increment curve prediction. The method comprises first obtaining local charging data of the target battery, including voltage and capacity, and performing discretization processing to obtain pre-processed data. Then, a voltage-capacity curve prediction model based on a convolutional neural network is used to predict the complete charging curve of the pre-processed data, and a capacity increment curve is generated. Next, characteristic data of the capacity increment curve, including the left peak height, the right peak height and the corresponding voltage, are extracted as clustering indicators for screening retired batteries. Finally, the discrimination result of the target battery is determined based on the characteristic data to achieve accurate screening and evaluation of the battery. The present invention can significantly shorten the time for acquiring clustering features, improve clustering efficiency, and achieve rapid screening of retired batteries. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0030] Figure 1 It is a schematic flow chart of a retired battery screening method based on capacity increment curve prediction of the present invention;

[0031] Figure 2 Graph showing the VQ curve and the corresponding IC curve in this embodiment;

[0032] Figure 3 This is a graph showing the attenuation of the peak value of the IC curve as the battery ages in this embodiment;

[0033] Figure 4 Schematic diagram of the prediction results and RMSE results in this embodiment; wherein, part (a) is a schematic diagram of 1000 cycles; part (b) is a schematic diagram of 2000 cycles; part (c) is a schematic diagram of 3000 cycles; part (d) is a schematic diagram of 4000 cycles; part (e) is a schematic diagram of 5000 cycles; part (f) is a schematic diagram of 6000 cycles; and part (g) is a schematic diagram of 7000 cycles. DETAILED DESCRIPTION

[0034] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0035] The purpose of the present invention is to provide a retired battery screening method, system, device and medium based on capacity increment curve prediction, which can greatly shorten the clustering feature acquisition time, improve clustering efficiency, and realize rapid screening of retired batteries.

[0036] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0037] like Figure 1 As shown, the present invention provides a retired battery screening method based on capacity increment curve prediction, comprising:

[0038] Step 100: Acquire local charging data of a target battery; the local charging data includes voltage and capacity.

[0039] Step 200: discretize the charging data to obtain pre-processed data.

[0040] Step 300: Input the preprocessed data into a voltage-capacity curve prediction model to predict a complete charging curve, and obtain a capacity increment curve based on the predicted complete charging curve; the voltage-capacity curve prediction model is constructed based on a convolutional neural network.

[0041] Step 400: extracting characteristic data of the capacity increment curve, and using the characteristic data as a clustering index to screen retired batteries, and determining the discrimination result of the target battery; the characteristic data includes a left peak height, a right peak height and a corresponding voltage.

[0042] Based on the above technical solution, the following embodiments are provided.

[0043] First, a retired battery screening method based on the capacity increment curve is determined.

[0044] When the same type of lithium batteries with different aging degrees are charged with the same amount of electricity, the battery with lower capacity decay has a smaller voltage rise value; the battery with higher capacity decay has a larger voltage rise value. Therefore, the voltage-capacity (VQ) curve can reflect the aging of the battery and the changes in characteristics such as capacity and internal resistance. However, the characteristics of the voltage-capacity curve are not obvious, especially its voltage platform is difficult to use directly, so it is necessary to convert it into a capacity increment curve with obvious peak characteristics, such as Figure 2 Among them, IC B Indicates the height of the peak on the left side of the IC curve, IC A represents the height of the right peak, V A is the voltage value corresponding to the peak position on the right.

[0045] The capacity increment curve is obtained by taking the first-order derivative (dQ / dV) of the battery voltage-capacity curve under constant current charging or discharging, which indicates the change in battery capacity corresponding to a unit voltage change during charging or discharging. Since the voltage during the plateau period of the VQ curve remains basically unchanged while the capacity increases rapidly, it is converted into a capacity increment curve peak. In practical applications, the data is discontinuous due to the influence of the sampling interval, so ΔQ / ΔV is usually used instead of the derivative. When ΔV is small enough, we have:

[0046]

[0047] As the battery ages, the capacity increment curve is significantly affected, such as Figure 3 .

[0048] As the number of battery cycles increases, the capacity increment curve peak height gradually decreases and the peak position gradually moves to the right, that is, the capacity increment peak height and peak position have a strong correlation with the battery aging degree. In summary, the peak height, peak position, capacity, and internal resistance are linearly correlated, which can reflect the changes in battery capacity and internal resistance as the aging degree deepens. Therefore, the peak height IC on the left side of the capacity increment curve is B , right peak height IC A And the corresponding voltage V A as a clustering indicator.

[0049] Then, a capacity increment curve prediction based on convolutional neural network is constructed.

[0050] The method based on the capacity increment curve solves the problem of difficulty in extracting aging features of the constant current and constant voltage charging VQ curve, but the acquisition of the capacity increment curve still requires the battery to be fully charged and discharged, and does not change the problem of long acquisition time. Therefore, this implementation method is based on a convolutional neural network, predicts the complete VQ curve through the local charging curve and derives the capacity increment curve, which greatly shortens the time required to obtain the capacity increment curve.

[0051] Data preprocessing: During the constant current charging process of lithium batteries, the battery capacity Q is regarded as a function of voltage:

[0052]

[0053] Among them, V low Indicates the starting voltage of lithium battery charging, V up is the upper limit voltage, Q(V) represents the constant current charging curve. Q(V) takes a value at V, where V is discretized at intervals of Δu (Δu is taken as 0.01V in this paper), as shown in formula (3). Therefore, the VQ curve of constant current charging can be saved as a 2×(N+1) matrix with V in the first row and Q(V) in the second row, where N is the number of samples of the discretized curve, as shown in formula (4).

[0054] In order to solve the problem that the starting and ending voltages of lithium battery charging are not fixed and the charging process is incomplete in actual use scenarios, this embodiment extracts a local segment of the charging curve with a length of m as the input matrix I for CNN training, see formula (5), where (1≤m≤N-1,2≤t≤N+1-m). The input matrix I of CNN includes two rows, V and Q(V), where the sequence V of equal step length represents the sampling position of Q(V) after discretization, the sequence V of different curves is the same, and Q(V) will vary between different batteries and different curves due to aging. Formula (5) sets the initial capacity value to zero by subtracting the capacity Q corresponding to the moment V(t-1) from the capacity sequence in I, and starts calculating the charging capacity until the moment V(t+m). Therefore, there is no need to rely on the complete charging process. The input matrix can be extracted from any voltage starting point, and the complete Q(V) sequence can be predicted with the help of the trained CNN, so as to achieve the prediction and output of the complete VQ curve and IC curve with a local charging segment of length (Nm).

[0055] V=[V low , V low +Δu, V low +2Δu, … , V low +NΔu] (3)

[0056] N=(V up -V low ) / Δu (4)

[0057]

[0058] In order to eliminate the influence of different dimensions of training data, data of different scales are unified to the same level. Therefore, normalization is required before inputting training data into CNN, as shown in formula (6):

[0059]

[0060] Among them, x norm represents the normalized training set; x is the original data, that is, the V and Q sequences before normalization; μ is the original mean of V and Q; σ is the original standard deviation of V and Q. The normalized training data has a mean of 0 and a variance of 1, which eliminates the physical meaning of the data and ensures the comparability of the data.

[0061] Network structure: The CNN structure and parameters used in this embodiment are shown in Table 1.

[0062] Table 1 CNN structure used for prediction

[0063]

[0064] Among them, the first layer of CNN is the input layer, which receives the processed data; the second layer is the convolution layer, which contains two filters, each filter contains 16 convolution kernels with a structure of 3×3, and the activation function is ReLU; the third layer is the pooling layer, the size of the pooling window is 3×3, and the pooling method is maximum pooling; the fourth, fifth, sixth and seventh layers are also convolutional layers and pooling layers, and the principles are the same as the second and third layers; the eighth layer is a fully connected layer (Dense) with 140 neuron nodes, and the activation function is ReLU; the ninth layer is the Dropout layer with a dropout rate of 0.2, which is used to randomly select neurons to zero to suppress overfitting; the tenth layer is the output layer, which has a total of 140 neurons, corresponding to 140 sampling points of the complete curve, and is used to output the predicted complete charging curve.

[0065] The overall prediction and clustering process of this implementation is as follows:

[0066] (1) Charging data processing: extract the voltage and capacity data of lithium battery constant current charging and perform discretization processing as CNN training data;

[0067] (2) Establish a CNN according to the previous content, and set appropriate parameters for different layers of CNN, such as input, convolution and pooling layers, and adjust the weight, batch size, number of cycles, etc. of CNN;

[0068] (3) Training CNN: Split the training data into a training set and a validation set, and shuffle the training data to prevent the model from overfitting. Train the CNN as required until its prediction accuracy meets actual needs.

[0069] (4) Capacity increment curve prediction: The discretized lithium battery constant current charging voltage and capacity segments are used as the input of CNN to predict the complete curve.

[0070] (5) Clustering based on capacity increment curve features: Extracting the IC of each battery A 、IC B 、V A Cluster the features and use t-test to eliminate outliers in the clustering results.

[0071] Example 1

[0072] The battery data uses the Oxford battery aging dataset, which includes 8 soft-pack lithium cobalt oxide batteries with a rated capacity of 0.74Ah produced by Kokam. The batteries are discharged at a constant current of 2C at a constant temperature of 40°C, and then charged at a constant current of 1C. The battery capacity is tested after every 100 charge and discharge cycles. The upper and lower limits of the battery voltage are 2.7V and 4.2V, respectively. According to the above method, the charging curve is discretized from 2.8V to 4.2V with a Δu interval of 0.01V as the training data and standardized. Batteries 1-6 are used as the training set, and the rest of the batteries are used as the test set.

[0073] For the test battery, a voltage window of 300mV length is used as the prediction input of CNN. From 0 to 7000 cycles, a total of 7 curves are extracted every 1000 cycles as verification data to prove that the proposed method is applicable to the complete life cycle of the battery. The root mean square error (RMSE) is used as the evaluation index of the charging curve prediction result, and the calculation formula (7) is:

[0074]

[0075] Where Q is the actual capacity sequence of the lithium battery, and Q' is the estimated result of the lithium battery capacity sequence. During the entire life cycle of the battery, the predicted results of the VQ curve and IC curve and the influence of different starting points of the input voltage window on the prediction error are shown in Figure 4 .

[0076] Figure 4The best and worst prediction results of the VQ curve and IC curve of each cycle, as well as the RMSE of the prediction curve, are affected by the different starting and ending positions of the prediction input voltage window. It can be seen that the proposed scheme can accurately predict the complete VQ curve and IC curve of the battery throughout its life cycle, and the maximum root mean square error of the worst prediction result is 14.21mAh, which is 1.92% of the rated capacity of the battery 740mAh. When the starting point of the voltage window input to CNN starts from the lower voltage limit, the prediction result of the IC curve is the worst and the peak error is large. This is because the voltage rises rapidly in a short time at the beginning of constant current charging, and the voltage window contains too little information due to the short time. In practical applications, if the battery starts charging from the cut-off voltage, the charging time is appropriately extended and the starting point of the voltage window is moved to the right to avoid this problem. When the starting point of the voltage window input to CNN is in the voltage range of 3.5V-3.9V, the RMSE of the prediction result is significantly reduced, which corresponds to the peak voltage range of the IC curve, reflecting the reference value of the peak of the IC curve. In addition, the downward trend of RMSE with the rightward shift of the voltage window at 6000 and 7000 cycles is not obvious in the first half. This is because the battery aging is too deep, causing the B peak of the IC curve to almost disappear. However, the battery capacity at this time has decayed to about 75% of the rated capacity, which is far below the 80% retirement standard. In practice, the capacity of the battery is usually greater than 80% when it is retired, so this problem does not exist in practice. At the starting voltage of 2.8V, the duration of the 300mV voltage window is as short as 9.8 minutes. Compared with the complete 56-minute 1C charging process, CNN shortens the curve acquisition time by about 82%. However, the 300mV voltage window corresponding to the IC peak lasts for 34 minutes because it includes the charging platform period, and CNN shortens the curve acquisition time by 40%. Even if the voltage window does not include the peak interval, the proposed CNN can still maintain a high prediction accuracy.

[0077] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0078] This article uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are only used to help understand the core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A retired battery screening method based on capacity increment curve prediction, characterized in that: include: Obtain local charging data of the target battery; The local charging data includes voltage and capacity; Discretizing the charging data to obtain preprocessed data; Inputting the preprocessed data into a voltage-capacity curve prediction model to predict a complete charging curve, and obtaining a capacity increment curve based on the predicted complete charging curve; the voltage-capacity curve prediction model is constructed based on a convolutional neural network; Extract characteristic data of the capacity increment curve, and use the characteristic data as a clustering index to screen retired batteries to determine the discrimination result of the target battery; the characteristic data includes a left peak height, a right peak height and a corresponding voltage.

2. The retired battery screening method based on capacity increment curve prediction according to claim 1 is characterized in that: The voltage-capacity curve prediction model consists of a ten-layer network structure, wherein the first layer is the input layer, which receives the discretized data; the second layer is the convolution layer, which contains two filters, each filter contains 16 convolution kernels with a structure of 3×3, and the activation function is ReLU; the third layer is the pooling layer, the size of the pooling window is 3×3, and the pooling method is maximum pooling; the fourth, fifth, sixth and seventh layers are also convolution layers and pooling layers, which are the same as the second and third layers; the eighth layer is a fully connected layer with 140 neuron nodes, and the activation function is ReLU; the ninth layer is the Dropout layer, and the drop rate is 0.2, which is used to randomly select neurons to zero to suppress overfitting; the tenth layer is the output layer, which has a total of 140 neurons, corresponding to 140 sampling points of the complete curve, and is used to output the predicted complete charging curve.

3. The retired battery screening method based on capacity increment curve prediction according to claim 1 is characterized in that: The pre-processed data is input into the voltage-capacity curve prediction model to predict the complete charging curve. The specific process includes: The preprocessed data is used as the input matrix I of CNN, and the input matrix I includes two rows of voltage and capacity, wherein the sequence voltage with equal step size represents the sampling position of the discretized capacity, the sequence voltage of different curves is the same, and the capacity is different between different batteries and different curves due to aging. The input matrix I is obtained by subtracting the capacity corresponding to the moment V(t-1) from the capacity sequence, setting the initial capacity value to zero, and starting to calculate the charging capacity until the moment V(t+m), 1≤m≤N-1, 2≤t≤N+1-m, N represents the total length of the charging curve, and m represents the length of the local charging curve.

4. The retired battery screening method based on capacity increment curve prediction according to claim 3 is characterized in that: The input matrix I is expressed as: Among them, V represents voltage and Q represents capacity.

5. The retired battery screening method based on capacity increment curve prediction according to claim 1 is characterized in that: Extracting characteristic data of the capacity increment curve, and using the characteristic data as a clustering index to screen retired batteries, and determining the discrimination result of the target battery, specifically includes: Extracting features from the capacity increment curve to obtain feature data corresponding to the curve; Clustering is performed based on the characteristic data, and combined with a t-test, outliers in the clustering results are eliminated to determine the discrimination results of the target battery and complete the retired battery screening.

6. A retired battery screening system based on capacity increment curve prediction, characterized in that: include: A data acquisition unit, used to obtain local charging data of a target battery; The local charging data includes voltage and capacity; A preprocessing unit, used for discretizing the charging data to obtain preprocessed data; A model prediction unit, used for inputting the preprocessed data into a voltage-capacity curve prediction model to predict a complete charging curve, and obtaining a capacity increment curve based on the predicted complete charging curve; the voltage-capacity curve prediction model is constructed based on a convolutional neural network; The retired battery screening unit is used to extract the characteristic data of the capacity increment curve, and use the characteristic data as a clustering index to screen the retired batteries to determine the discrimination result of the target battery; the characteristic data includes the left peak height, the right peak height and the corresponding voltage.

7. An electronic device, characterized in that: The electronic device comprises a memory and a processor, wherein the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute the retired battery screening method based on capacity increment curve prediction according to any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that: It stores a computer program, which, when executed by a processor, implements the retired battery screening method based on capacity increment curve prediction as described in any one of claims 1 to 5.

Images